As there are couple of updating of SDK , and my build environment mix up lots of different version of CUDA/CUDNN on same box, in order to better control those SDK version , here
is docker image build example for latest NCSDK-1.12.00.01, verified on ubuntu x86-64.
If lots of people like this, i can create a pull request to see it can be accepted.
I would like you to suggest this article.
Intel seems to be striving to improve the installation process.
Yes, I saw it. However due to my environments had different DL framework and using py virtualenv to control version of python package dependency, besides new framework likes TF request different CUDA, CUDNN version than previous version.
Let's why I prefer to pickup docker as devops env, I can free switch to any DL framework without impact performance lost on GPU access vs VM case , all still in same server.